Anthropic's Mythos AI model isn't the threat you think

When Anthropic launched its powerhouse AI model, Mythos, last month, it triggered immediate panic. The company warned the model had already uncovered thousands of software vulnerabilities across major browsers and operating systems, sparking fears of a looming, AI-driven hacking wave. The White House even began debating strict new release rules.

But one month later, cybersecurity veterans are urging everyone to take a deep breath. Inside the industry, the consensus is clear: the threat of Mythos unleashing a wave of untamed cyberattacks has been wildly overstated, News.Az reports, citing Reuters.

"There’s a really big communication gap between practitioners and policymakers," explains Isaac Evans, CEO of software security firm Semgrep. While Mythos is a genuine technical leap, experts say the political panic doesn't align with how hacking actually works in the real world.

The truth is, finding software bugs with artificial intelligence isn't a new breakthrough. "We’ve been able to use AI to find more bugs than we know what to do with for months, if not years," noted one vulnerability researcher.

Instead, the real bottleneck for both hackers and defenders isn't finding the flaws—it's validating them, prioritizing them, and fixing them without crashing the entire system. Mythos makes finding vulnerabilities easier with simpler prompts, but it doesn't automatically grant amateur hackers elite capabilities.

"If you have a Formula One car but you've only ever driven a bike, you might be able to get it to go straight," says Anthony Grieco, Cisco’s chief security and trust officer. "But you're not going to maximize the track time out of the gate."

To actually weaponize or defend with a model like Mythos, organizations need massive computing power and highly specialized environments. For now, these infrastructure demands keep the model out of reach for average bad actors.

Furthermore, today's cybercriminals are already devastatingly effective without AI. Former FBI cybersecurity official Cynthia Kaiser points out that ransomware attacks currently execute in under an hour using traditional methods.

While Anthropic is currently working with the U.S. government to manage access through defensive testing initiatives like Project Glasswing, tech experts emphasize that Mythos is an evolution, not a cyber-apocalypse. The barrier to entry for advanced AI is lowering, but for now, human expertise remains the ultimate bottleneck.

News.Az 

By Aysel Mammadzada